Check extraction

Prepare data

We load the the reported hospitalization data on data.gouv, on this page.

Here are the metadata

#metadonnees-covid-hospit-incid-reg.csv
metadata <- read_csv2(
  "data/0_metadata/metadonnees-covid-hospit-incid-reg.csv",
  locale=locale(encoding="latin1")
  )

#metadonnees-donnees-hospitalieres-covid19-maj.csv
metadata <- read_csv2(
  "data/0_metadata/metadonnees-donnees-hospitalieres-covid19-maj20210810.csv",
  locale=locale(encoding="latin1")
  )

#metadonnees-donnees-hospitalieres-covid19-nouveaux.csv
metadata <- read_csv2(
  "data/0_metadata/metadonnees-hospit-incid.csv",
  locale=locale(encoding="latin1")
  )

#metadonnees-donnees-hospitalieres-covid19-classes-age-maj.csv
metadata <- read_csv2(
  "data/0_metadata/metadonnees-donnees-hospitalieres-covid19-classes-age-maj20210810.csv",
  locale=locale(encoding="latin1")
  )

metadonnees-covid-hospit-incid-reg.csv : rea admissions, region

metadonnees-donnees-hospitalieres-covid19-maj.csv : dep, sex, hosp, rea, dc, hospconv

metadonnees-donnees-hospitalieres-covid19-nouveaux.csv : departement new (incid) hosp, rea, dc

metadonnees-donnees-hospitalieres-covid19-classes-age-maj.csv : region, age, hosp, rea, dec, hospconv

metadonnees-services-hospitaliers-covid19.csv => not used

metadonnees-sexe.csv pas important => not used (0 both, 1 men, 2 women)

file

loaded

admissions : covid-hospit-incid.csv (permalien https://www.data.gouv.fr/fr/datasets/r/6fadff46-9efd-4c53-942a-54aca783c30c)

A charger !!!

covid-hosp-ad-age https://www.data.gouv.fr/fr/datasets/r/dc7663c7-5da9-4765-a98b-ba4bc9de9079 by region, by age, admissions to hospital

not loaded :

covid-hosp-txad-age-fra : reported to population, whole france, by age, deaths, hosp, SC

covid-hosp-txad-reg: reported to population, by region, by age, deaths, hosp, SC

covid-hosp-txad-fra : reported to population, whole france, deaths, hosp, SC

covid-hospit-incid-reg : new rea by region

covid-hospit-clage : region x age, for hosp, rea and hospConv

covid-hospit-etab : not interesting

data_gouv_new_hosp_rea <- 
  read.csv(url("https://www.data.gouv.fr/fr/datasets/r/6fadff46-9efd-4c53-942a-54aca783c30c"), sep=";") %>%
  mutate(date = as.Date(jour))
data_gouv_beds_hosp_rea <- 
  read.csv(url("https://www.data.gouv.fr/fr/datasets/r/63352e38-d353-4b54-bfd1-f1b3ee1cabd7"), sep=";")  %>%
  mutate(date = as.Date(jour))

#critical car beds, hospitlization beds and conventional hospitalization beds
true_data_beds_hosp_rea <- data_gouv_beds_hosp_rea %>%
  filter(sexe =="0", # 0 = men + women, 1 = men, 2 = women
         #enlève l'Outre-Mer car les scénarios de Pasteur uniquement pour la France Métropolitaine
         dep != 971 & dep != 972 & dep != 973 & dep != 974 & dep != 976 & dep != 978) %>%
  group_by(date) %>% #grouper tous les départements ensembles
  dplyr::summarise(hosp = sum(hosp, na.rm = T), 
            rea = sum(rea, na.rm = T),
            HospConv = sum(HospConv, na.rm = T))

#pour les echos pasteur 29 avril ile de france
true_data_beds_hosp_rea_IDF <- data_gouv_beds_hosp_rea %>%
  filter(sexe =="0", # 0 = hommes + femmes, 1=hommes, 2=femmes
         #enlève l'Outre-Mer car les scénarios de Pasteur uniquement pour la France Métropolitaine
         dep == 75 | dep == 92 | dep == 93 | dep == 94 | dep == 91 | dep == 95 | dep == 78 | dep == 77) %>% 
  group_by(date) %>% #grouper tous les départements ensembles
  dplyr::summarise(hosp = sum(hosp, na.rm = T), 
            rea = sum(rea, na.rm = T),
            HospConv = sum(HospConv, na.rm = T))

#Nouvelles admissions à l'hôpital et en réanimation (moyenné sur 7 jours)
true_data_new_hosp_rea <- data_gouv_new_hosp_rea %>% 
  filter( #enlève l'Outre-Mer car les scénarios de Pasteur uniquement pour la France Métropolitaine
         dep != 971 & dep != 972 & dep != 973 & dep != 974 & dep != 976 & dep != 978) %>% 
  group_by(date) %>% #grouper tous les départements ensembles
  dplyr::summarise(incid_hosp = sum(incid_hosp, na.rm = T), 
            incid_rea = sum(incid_rea, na.rm = T)) %>%
  mutate(new_rea_right = rollmean(incid_rea, 7, na.pad = T, align = "right"),#mean of 7 last days
         new_hosp_right = rollmean(incid_hosp, 7, na.pad = T, align = "right"),
         new_rea_center = rollmean(incid_rea, 7, na.pad = T, align = "center"),#centered mean
         new_hosp_center = rollmean(incid_hosp, 7, na.pad = T, align = "center"))

true_data_new_hosp_rea_no_mean <- data_gouv_new_hosp_rea %>% 
  filter( #enlève l'Outre-Mer car les scénarios de Pasteur uniquement pour la France Métropolitaine
    dep != 971 & dep != 972 & dep != 973 & dep != 974 & dep != 976 & dep != 978) %>% 
  group_by(date) %>% #grouper tous les départements ensembles
  dplyr::summarise(new_hosp = sum(incid_hosp, na.rm = T), 
            new_rea = sum(incid_rea, na.rm = T))

#Pour l'INSERM : admissions hebdomadaire à l'hôpital
true_data_new_hosp_rea_weekly <- data_gouv_new_hosp_rea %>% 
  group_by(date) %>% #je n'ai pas filtré les outre-mer ici car pas précisé, mais ne change pas grand-chose
  dplyr::summarise(incid_hosp = sum(incid_hosp, na.rm = T)) %>%
  mutate(new_hosp_week = rollsum(incid_hosp, 7, na.pad = T, align = "left")) #patients arrivés dans les 7 derniers jours)

Paireau et al paper

true_data_Paireau_et_al <- readRDS("Data/full_data.rds") %>%
  filter(region == "metropolitan") %>%
  select(date, iHosp, iHosp_smooth, iICU, iICU_smooth, inHosp, inHosp_smooth, inICU, inICU_smooth) %>%
  distinct()

#"smooth" data reported multiple time for each date, so we synthesize it
true_data_Paireau_et_al <- true_data_Paireau_et_al %>%
  group_by(date) %>%
  summarise_all(mean, na.rm=T) %>%
  mutate_all(round, 0)

for INSERM data (hospital admissions by week)

remove the overseas regions, see this map https://en.wikipedia.org/wiki/INSEE_code#/media/File:Carte-des-codes-des-regions-selon-l-INSEE.jpg

temp <- 
  read_csv2(url("https://www.data.gouv.fr/fr/datasets/r/dc7663c7-5da9-4765-a98b-ba4bc9de9079")) %>%
  filter(!reg %in% c("01", "02", "03", "04", "06"))

INSERM_true <- temp %>% filter(cl_age90 == "0") %>%
  rename(date = Semaine) %>%
  group_by(date) %>%
  summarise(
    NewAdmHospit = sum(NewAdmHospit, na.rm=T)
  ) 

INSERM_true <- temp %>% filter(cl_age90 != "0") %>%
  rename(date = Semaine) %>%
  group_by(date) %>%
  summarise(
    NewAdmHospit = sum(NewAdmHospit, na.rm=T)
  )


INSERM_true$date <- gsub("S", "W", INSERM_true$date)
INSERM_true$date <- paste0(INSERM_true$date, "-4")
INSERM_true$date <- ISOweek2date(INSERM_true$date)

#remplacer par ce qu'il y a en dessous pour calculer min, 
#puis inner join avec fichier real pour garder que les dates qui matchent (important pour INSERM)
f_compute_error <- function(date_begin, date_end, dataset){
  #dates delimitating the comparison period
  date_min <- as.Date(date_begin)
  date_max <- as.Date(date_end)
  #preparing file : gets date and true data value on the period
  reality_file <- dataset %>%
    select(date, reality) %>%
    filter(date>date_min & date<date_max)
  #computin min, med and max of scenarios
  temp <- dataset %>% 
    select(-reality, -reality_report) %>%
    rowwise() %>%
    mutate(
      min = min(c_across(-date), na.rm=T),
      med = median(c_across(-date), na.rm=T),
      max = max(c_across(-date), na.rm=T),
    )
  #joins 2 files
  reality_file <- inner_join(reality_file, temp, by="date")
  reality_file <- reality_file %>%
    mutate(
      error_min = round((min/reality)*100),
      error_med = round((med/reality)*100),
      error_max = round((max/reality)*100)
    )
  return (reality_file)
}

f_graph <- 
  function(
    true_data, scenarios, variable, 
    x_label_publication, y_label_publication,
    x_min, x_max, y_max,
    str_y, str_reality
  ){
    modellers_true_data <- scenarios %>%
      select(date, reality)
    
    scenarios <- scenarios %>%
      select(-reality) %>%
      gather(key=scenario, value = value, -date)

    
    p <- ggplot(data = scenarios) + 
      #scenarios lines
      geom_line(
        aes(
          x=date, y=value, 
          group=scenario, color="scenarios"
        ),
        size = 1
      ) + 
      #reality line
      geom_line(
        data= true_data, 
        aes(
          x=date, y=!!as.symbol(variable), 
          color = str_reality
          ),
        size = 1
        ) +
      #modellers reality line
      geom_point(
        data = modellers_true_data,
        aes(
          date, reality, color = "reality in report"
          ) 
      ) +
      #publication date line and label
      geom_vline(
        xintercept = as.Date(x_label_publication), linetype="dashed"
      ) +
      annotate(
        'text', x = as.Date(x_label_publication)-1, y = y_label_publication, label = "publication\ndate", 
        color = "black", fontface = "italic", family = "Times New Roman", hjust=1
      ) +
      # x and y limits
      xlim(as.Date(x_min), as.Date(x_max)) + ylim(0, y_max) + 
      g_theme +
      labs(
        title = "",
        subtitle = "",
        caption = Source,
        x="", y= str_y
        )
    
    return(p)
  }

function to correct curves

f_offset <- function(dataset_scenarios, dataset_reality, variable_select){
  #reality 
  temp_reality <- dataset_reality %>% select(date, reality = !!as.symbol(variable_select))
  temp_reality$date <- temp_reality$date + x_reality_offset #on date (x)
  temp_reality$reality <- temp_reality$reality + y_reality_offset #on values (y)
  
  #reality in report
  temp_reality_report <- dataset_scenarios %>% select(date, reality_report = reality)
  
  
  #scenarios offset
  temp_scenarios <- dataset_scenarios %>% select(-reality)
  temp_scenarios$date <- temp_scenarios$date + x_scenarios_offset #on dates (x)
  temp_scenarios[,-1] <- lapply(temp_scenarios[,-1], function(x) x+y_scenarios_offset) #on values (y)
  
  temp <- full_join(temp_scenarios, temp_reality_report, by="date")
  temp <- left_join(temp, temp_reality)
  
  return(temp)
}

graph of corrected data

f_graph_corrected <- 
  function(
    scenarios, 
    x_label_publication, y_label_publication,
    x_min, x_max, y_max,
    str_y, str_reality
  ){
    modellers_true_data <- scenarios %>%
      select(date, reality_report)
    
    true_data <- scenarios %>%
      select(date, reality)
    
    scenarios <- scenarios %>%
      select(-reality, -reality_report) %>%
      gather(key=scenario, value = value, -date)

    
    p <- ggplot(data = scenarios) + 
      #scenarios lines
      geom_line(
        aes(
          x=date, y=value, 
          group=scenario, color="scenarios"
        ),
        size = 1
      ) + 
      #reality line
      geom_line(
        data= true_data, 
        aes(
          x=date, y=reality, 
          color = str_reality
          ),
        size = 1
        ) +
      #modellers reality line
      geom_point(
        data = modellers_true_data,
        aes(
          date, reality_report, color = "reality in report"
          ) 
      ) +
      #publication date line and label
      geom_vline(
        xintercept = as.Date(x_label_publication), linetype="dashed"
      ) +
      annotate(
        'text', x = as.Date(x_label_publication)-1, y = y_label_publication, label = "publication\ndate", 
        color = "black", fontface = "italic", family = "Times New Roman", hjust=1
      ) +
      # x and y limits
      xlim(as.Date(x_min), as.Date(x_max)) + ylim(0, y_max) + 
      g_theme +
      labs(
        title = "",
        subtitle = "",
        caption = Source,
        x="", y= str_y
        )
    
    return(p)
  }

graph of errors

f_graph_error <- function(
    dataset, x_label, y_label
    ){
  ggplot(dataset, aes(date)) + 
  geom_line(aes(y=error_med)) +
  geom_ribbon(aes(ymin = error_min, ymax = error_max), alpha = 0.1) + 
  geom_hline(yintercept = 100) +
  geom_vline(xintercept=as.Date(x_label), linetype="dashed") +
  annotate(
    'text', x = as.Date(x_label)-1, y = y_label, label = "publication\ndate", 
    color = "black", fontface = "italic", family = "Times New Roman", hjust=1
  ) +
  ylim(0, NA) +
  labs(
    x="", y="% of reality value",
    title = "Median, min and max relative errors of scenarios vs reality",
    subtitle = "line: median scenario ; area: min and max scenarios"
  )
}

April 28, 2020

Source: Les Echos newspaper, April 29, 2020. Specified “dated Tuesday”, so we deduce that the publication date from Tuesday April 28. Identified by Google search.

Original

Reproduced

There is a slight horizontal offset between the curves, already present in the original figure. To match the report reality, we offset the Paireau et al. reality curve by -1 day, and the scenarios reality curves by +1 day.

#offset values
x_reality_offset <- -1
y_reality_offset <- 0
x_scenarios_offset <- 1
y_scenarios_offset <- 0

Before correction

scenarios <- read_csv("data/2020_04_29/beds_rea.csv") %>%
  mutate(date = as.Date(date, format = "%Y/%m/%d", optional = T))

f_graph(
  true_data_beds_hosp_rea_IDF, scenarios, 
  "rea", 
  "2020-04-28", 1000, #publication date label
  "2020-03-19", "2020-06-28", #date limits
  NA, # y limits
  "ICU beds in Ile-de-France", #y axis label
  "reality in data.gouv" #reality label
)

After correction

temp <- f_offset(scenarios, true_data_beds_hosp_rea_IDF, "rea")

f_graph_corrected(
  temp, 
  "2020-04-28", 1000, #publication date label
  "2020-03-19", "2020-06-28", #date limits
  NA, # y limits
  "Intensive care beds in Ile-de-France", 
  "reality in data.gouv"
)

write_csv(temp, "data/2020_04_29/corrected_scenarios.csv")

Error

error <- f_compute_error("2020-03-29", "2020-06-28", temp)

f_graph_error(
  error,
  "2020-04-28", 300 #publication date label
  )

October 30, 2020

Source: Les Echos newspaper, November 3, 2020. We know the publication date from the statement “The scientists from Pasteur Institute and Santé Publique France updated their epidemic scenarios on October 30”. Identified by Google search.

checker la sortie du confinement

Orginal

Reproduced

There is no need to correct the data.

#offset values
x_reality_offset <- 0
y_reality_offset <- 0
x_scenarios_offset <- 0
y_scenarios_offset <- 0

Before correction

scenarios <- read_csv("data/2020_10_30/beds_rea.csv") %>%
  mutate(date = as.Date(date, format = "%Y/%m/%d", optional = T))

f_graph(
  true_data_Paireau_et_al, scenarios, 
  "inICU", 
  "2020-10-30", 5000, #publication date label
  "2020-10-01", "2020-12-15", #date limits
  NA, # y limits
  "ICU beds",
  "reality in Paireau et al."
)

After correction

temp <- f_offset(
  scenarios, 
  true_data_Paireau_et_al %>% select(date, inICU), 
  "inICU"
  )

f_graph_corrected(
  temp, 
  "2020-10-30", 5000, #publication date label
  "2020-10-01", "2020-12-15", #date limits
  NA, # y limits
  "ICU beds", 
  "reality in data.gouv"
)

write_csv(temp, "data/2020_10_30/corrected_scenarios.csv")

Error

error <- f_compute_error("2020-10-01", "2020-12-15", temp)

f_graph_error(
  error,
  "2020-10-30", 200 #publication date label
  )

January 16, 2021

Source: INSERM/Pasteur report, January 16, 2021. Cited in the January 29, 2021 report, which was identified on Pasteur Institute’s website, on this page.

Original

Fig 1

Reproduced

We do not correct the data

#offset values
x_reality_offset <- 0
y_reality_offset <- 0
x_scenarios_offset <- 0
y_scenarios_offset <- 0

Before correction

scenarios <- read_csv("data/2021_01_16/weekly_hospital.csv")
scenarios$date <- paste0(scenarios$date, "-4")
scenarios$date <- ISOweek2date(scenarios$date)

temp2 <- true_data_Paireau_et_al %>%
  select(date,iHosp) %>%
  mutate(
    date = ISOweek(date),
    date = ISOweek2date(paste0(date, "-4"))
    ) %>%
  group_by(date) %>%
  summarise(iHosp = sum(iHosp, na.rm=T)) 

f_graph(
  temp2, scenarios, 
  "iHosp", 
  "2021-01-16", 1000, #publication date label
  "2020-10-01", "2021-05-01", #date limits
  NA, # y limits
  "weekly hospital admissions",
  "reality in Paireau et al."
)

  #confinement de 16 départements le 20 mars cf https://fr.wikipedia.org/wiki/Chronologie_de_la_pand%C3%A9mie_de_Covid-19_en_France
  #2 semaines pour voir les effets sur hospitalisations. correspond aussi au confinement général du 3 avril

After correction

temp <- f_offset(
  scenarios, 
  temp2, 
  "iHosp"
  )

f_graph_corrected(
  temp, 
  "2021-01-16", 1000, #publication date label
  "2020-10-01", "2021-05-01", #date limits
  NA, # y limits
  "weekly hospital admissions",
  "reality in Paireau et al."
)

write_csv(temp, "data/2021_01_16/corrected_scenarios.csv")

error

error <- f_compute_error("2020-10-01", "2021-05-01", temp)

f_graph_error(
  error,
  "2021-01-16", 250 #publication date label
  )

February 2, 2021

Source: INSERM/Pasteur report, February 2, 2021. Identified on Pasteur Institute’s website, on this page.

Original

Fig 2

Reproduced

We do not correct the data

#offset values
x_reality_offset <- 0
y_reality_offset <- 0
x_scenarios_offset <- 0
y_scenarios_offset <- 0

Before correction

scenarios <- read_csv("data/2021_02_02/new_hosp_INSERM.csv")
scenarios$date <- paste0(scenarios$date, "-4")
scenarios$date <- ISOweek2date(scenarios$date)


temp2 <- true_data_Paireau_et_al %>%
  select(date,iHosp) %>%
  mutate(
    date = ISOweek(date),
    date = ISOweek2date(paste0(date, "-4"))
    ) %>%
  group_by(date) %>%
  summarise(iHosp = sum(iHosp, na.rm=T)) 

f_graph(
  temp2, scenarios, 
  "iHosp", 
  "2021-02-02", 1000, #publication date label
  "2020-10-01", "2021-05-01", #date limits
  NA, # y limits
  "weekly hospital admissions",
  "reality in Paireau et al."
)

#confinement de 16 départements le 20 mars cf https://fr.wikipedia.org/wiki/Chronologie_de_la_pand%C3%A9mie_de_Covid-19_en_France
  #2 semaines pour voir les effets sur hospitalisations. correspond aussi au confinement général du 3 avril

After correction

temp <- f_offset(
  scenarios, 
  temp2, 
  "iHosp"
  )

f_graph_corrected(
  temp, 
  "2021-02-03", 1000, #publication date label
  "2020-10-01", "2021-05-01", #date limits
  NA, # y limits
  "weekly hospital admissions",
  "reality in Paireau et al."
)

write_csv(temp, "data/2021_02_02/corrected_scenarios.csv")

Error

error <- f_compute_error("2020-10-01", "2021-05-01", temp)

f_graph_error(
  error,
  "2021-02-02", 250 #publication date label
  )

February 8, 2021

Source: Pasteur report, February 8, 2021. Identified on Pasteur Institute’s website, on this page.

Original

probably something to adjust

Fig 2A

Fig 6A

Fig 7C

Reproduced

We do not correct the data, except for a small offset by 100 for reality

#offset values
x_reality_offset <- 0
y_reality_offset <- 100
x_scenarios_offset <- 0
y_scenarios_offset <- 0

Before correction

scenarios <- read_csv("data/2021_02_08/new_hosp.csv") %>%
  mutate(date = as.Date(date, format = "%Y/%m/%d", optional = T))

f_graph(
  true_data_Paireau_et_al, scenarios, 
  "iHosp_smooth", 
  "2021-02-08", 5000, #publication date label
  "2021-01-01", "2021-06-01", #date limits
  NA, # y limits
  "daily hospital admissions",
  "reality in Paireau et al."
) +
  geom_line(
    data=true_data_Paireau_et_al, aes(date, iHosp), alpha=.4, color="red"
  )

After correction

temp <- f_offset(
  scenarios, 
  true_data_Paireau_et_al %>% select(date, iHosp_smooth),
  "iHosp_smooth"
  )

f_graph_corrected(
  temp, 
  "2021-02-08", 5000, #publication date label
  "2021-01-01", "2021-06-01", #date limits
  NA, # y limits
  "weekly hospital admissions",
  "reality in Paireau et al."
)  +
  geom_line(
    data=true_data_Paireau_et_al, aes(date, iHosp+y_reality_offset), alpha=.4, color="red"
  )

write_csv(temp, "data/2021_02_08/corrected_scenarios.csv")

Error

error <- f_compute_error("2021-01-01", "2021-03-27", temp)

f_graph_error(
  error,
  "2021-02-08", 250 #publication date label
  )

February 23, 2021

Source: Pasteur report, February 23, 2021. Identified on Pasteur Institute’s website, on this page.

Original

apparemment mieux de prendre med = (max+min)/2 que le vrai med, pose problème quand bcp de scénarios multipliés

Fig 2C

Fig 5A, 5C and 5E

Reproduced

We do not correct the data

#offset values
x_reality_offset <- 0
y_reality_offset <- 0
x_scenarios_offset <- 0
y_scenarios_offset <- 0

Before correction

#February 23 2021
#besoin de réaligner leurs données sur la réalité (ne compte surement pas exactement la meme chose)
scenarios <- read_csv("data/2021_02_23/new_hosp.csv") %>%
  mutate(date = as.Date(date, format = "%Y/%m/%d", optional = T))

f_graph(
  true_data_Paireau_et_al, scenarios, 
  "iHosp_smooth", 
  "2021-02-23", 3000, #publication date label
  "2021-01-15", "2021-07-01", #date limits
  NA, # y limits
  "daily hospital admissions",
  "reality in Paireau et al."
) 

  #confinement de 16 départements le 20 mars cf https://fr.wikipedia.org/wiki/Chronologie_de_la_pand%C3%A9mie_de_Covid-19_en_France
  #2 semaines pour voir les effets sur hospitalisations. correspond aussi au confinement général du 3 avril

After correction

temp <- f_offset(
  scenarios, 
  true_data_Paireau_et_al %>% select(date, iHosp_smooth), 
  "iHosp_smooth"
  )

f_graph_corrected(
  temp, 
  "2021-02-23", 3000, #publication date label
  "2021-01-15", "2021-07-01", #date limits
  NA, # y limits
  "weekly hospital admissions",
  "reality in Paireau et al."
)

write_csv(temp, "data/2021_02_23/corrected_scenarios.csv")

Error

error <- f_compute_error("2021-01-16", "2021-03-27", temp)

f_graph_error(
  error,
  "2021-02-23", 250 #publication date label
  )

April 26, 2021

Source: Pasteur report, April 26, 2021. Identified on Pasteur Institute’s website, on this page.

Original

Fig 3B and 3D

Reproduced

Just a slight vertical correction on reality curve

#offset values
x_reality_offset <- 0
y_reality_offset <- 30
x_scenarios_offset <- 0
y_scenarios_offset <- 0

Before correction

scenarios <- read_csv("data/2021_04_26/new_hosp.csv") %>%
  mutate(date = as.Date(date, format = "%Y/%m/%d", optional = T))



f_graph(
  true_data_Paireau_et_al, scenarios, 
  "iHosp_smooth",
  "2021-04-26", 1000, #publication date label
  "2021-01-15", "2021-07-01", #date limits
  NA, # y limits
  "daily hospital admissions", 
  "reality in Paireau et al."
) +
  geom_line(
    data=true_data_Paireau_et_al, aes(date, iHosp), alpha=.4, color="red"
  )  +
  ylim(0, 2000)

After correction

temp <- f_offset(
  scenarios, 
  true_data_Paireau_et_al %>% select(date, iHosp_smooth), 
  "iHosp_smooth"
  )

f_graph_corrected(
  temp, 
  "2021-04-26", 1000, #publication date label
  "2021-01-15", "2021-07-01", #date limits
  NA, # y limits
  "daily hospital admissions",
  "reality in Paireau et al."
) +
  geom_line(
    data=true_data_Paireau_et_al, aes(date, iHosp), alpha=.4, color="red"
  ) +
  ylim(0, 2000)

write_csv(temp, "data/2021_04_26/corrected_scenarios.csv")

Error

error <- f_compute_error("2021-03-15", "2021-06-15", temp)

f_graph_error(
  error,
  "2021-04-26", 250 #publication date label
  )

May 21 Scenarios TBD

Source: Pasteur report, May 21, 2021. Identified on Pasteur Institute’s website, on this page.

Original

fig 3A

fig 3C

fig 3E

fig 3G

Reproduced

Just a small x offset by 1 day

#offset values
x_reality_offset <- 1
y_reality_offset <- 0
x_scenarios_offset <- 0
y_scenarios_offset <- 0

Before correction

scenarios <- read_csv("data/2021_05_21/SC_beds.csv") %>%
  mutate(date = as.Date(date, format = "%Y/%m/%d", optional = T))

f_graph(
  true_data_Paireau_et_al, scenarios, 
  "inICU",
  "2021-05-21", 1000, #publication date label
  "2021-01-15", "2021-07-01", #date limits
  NA, # y limits
  "ICU beds", 
  "reality in Paireau et al."
) 

After correction

temp <- f_offset(
  scenarios, 
  true_data_Paireau_et_al %>% select(date, inICU), 
  "inICU"
  )

f_graph_corrected(
  temp, 
  "2021-05-21", 1000, #publication date label
  "2021-01-15", "2021-07-01", #date limits
  NA, # y limits
  "ICU beds", 
  "reality in Paireau et al."
) 

write_csv(temp, "data/2021_05_21/corrected_scenarios.csv")

Error

error <- f_compute_error("2021-03-15", "2021-06-15", temp)

f_graph_error(
  error,
  "2021-05-21", 250 #publication date label
  )

July 26, 2021 REEXTRAIRE ?

Source: Pasteur report, July 26, 2021. Identified on Pasteur Institute’s website, on this page.

Original

Fig 6

zoomed

knitr::include_graphics("data/2021_07_26/fig6_zoom.png")

all

Fig 5

zoomed

knitr::include_graphics("data/2021_07_26/fig5_zoom.png")

all

Reproduced

Small vertical offset of -30

#offset values
x_reality_offset <- 0
y_reality_offset <- -30
x_scenarios_offset <- 0
y_scenarios_offset <- 0

Before correction

scenarios <- read_csv("data/2021_07_26/beds_SC.csv") %>%
  mutate(date = as.Date(date, format = "%Y/%m/%d", optional = T))

f_graph(
  true_data_beds_hosp_rea, scenarios, 
  "rea",
  "2021-07-26", 5000, #publication date label
  "2021-06-15", "2021-10-01", #date limits
  NA, # y limits
  "ICU beds", 
  "reality in data.gouv"
) 

After correction

temp <- f_offset(
  scenarios, 
  true_data_beds_hosp_rea, 
  "rea"
  )

f_graph_corrected(
  temp, 
  "2021-07-26", 5000, #publication date label
  "2021-06-15", "2021-10-01", #date limits
  NA, # y limits
  "ICU beds", 
  "reality in data.gouv"
) 

write_csv(temp, "data/2021_07_26/corrected_scenarios.csv")

Error

error <- f_compute_error("2021-07-15", "2021-10-10", temp)

f_graph_error(
  error,
  "2021-07-26", 250 #publication date label
  )

August 5, 2021

Source: Pasteur report, August 5, 2021. Identified on Pasteur Institute’s website, on this page.

Original

Fig F

Reproduced

We do not correct the data

#offset values
x_reality_offset <- -2
y_reality_offset <- -80
x_scenarios_offset <- 0
y_scenarios_offset <- 0

Before correction

scenarios <- read_csv("data/2021_08_05/beds_SC.csv") %>%
  mutate(date = as.Date(date, format = "%Y/%m/%d", optional = T))

f_graph(
  true_data_beds_hosp_rea, scenarios, 
  "rea",
  "2021-08-05", 5000, #publication date label
  "2021-06-15", "2021-10-01", #date limits
  NA, # y limits
  "ICU beds", 
  "reality in data.gouv"
) 

After correction

temp <- f_offset(
  scenarios, 
  true_data_beds_hosp_rea, 
  "rea"
  )

f_graph_corrected(
  temp, 
  "2021-08-05", 5000, #publication date label
  "2021-06-15", "2021-10-01", #date limits
  NA, # y limits
  "ICU beds", 
  "reality in data.gouv"
) 

write_csv(temp, "data/2021_08_05/corrected_scenarios.csv")

Error

error <- f_compute_error("2021-07-15", "2021-10-10", temp)

f_graph_error(
  error,
  "2021-08-05", 250 #publication date label
  )

October 4, 2021

Source: Pasteur report, October 4, 2021. Identified on Pasteur Institute’s website, on this page.

Original

Fig 9

Reproduced

Horizontal offset by -1 day, vertical offset by -40

#offset values
x_reality_offset <- -1
y_reality_offset <- -40
x_scenarios_offset <- 0
y_scenarios_offset <- 0

Before correction

scenarios <- read_csv("data/2021_10_04/new_hosp.csv") %>%
  mutate(date = as.Date(date, format = "%Y/%m/%d", optional = T))

f_graph(
  true_data_new_hosp_rea, scenarios, 
  "new_hosp_center",
  "2021-10-04", 1000, #publication date label
  "2021-07-01", "2022-01-01", #date limits
  NA, # y limits
  "daily hospital admissions beds", 
  "reality in data.gouv"
)

After correction

temp <- f_offset(
  scenarios, 
  true_data_new_hosp_rea, 
  "new_hosp_center"
  )

f_graph_corrected(
  temp, 
  "2021-10-04", 1000, #publication date label
  "2021-07-01", "2022-01-01", #date limits
  NA, # y limits
  "daily hospital admissions beds", 
  "reality in data.gouv"
) 

write_csv(temp, "data/2021_10_04/corrected_scenarios.csv")

Error

error <- f_compute_error("2021-09-15", "2021-12-20", temp)

f_graph_error(
  error,
  "2021-10-04", 250 #publication date label
  )

January 7, 2021

Source: Pasteur report, January 7, 2022. Identified on Pasteur Institute’s website, on this page.

Original

Figure 4

Reproduced

-2 day for reality, -# days for scenarios

#offset values
x_reality_offset <- -2
y_reality_offset <- 0
x_scenarios_offset <- -1
y_scenarios_offset <- 0

Before correction

scenarios <- read_csv("data/2022_01_07/beds_SC_low_VE.csv") %>%
  mutate(date = as.Date(date, format = "%Y/%m/%d", optional = T))


f_graph(
  true_data_beds_hosp_rea, scenarios, 
  "rea",
  "2022-01-07", 1000, #publication date label
  "2021-12-01", "2022-04-01", #date limits
  NA, # y limits
  "ICU beds", 
  "reality in data.gouv"
)

After correction

temp <- f_offset(
  scenarios, 
  true_data_beds_hosp_rea, 
  "rea"
  )

f_graph_corrected(
  temp, 
  "2022-01-07", 1000, #publication date label
  "2021-12-01", "2022-04-01", #date limits
  NA, # y limits
  "ICU beds", 
  "reality in data.gouv"
) 

write_csv(temp, "data/2022_01_07/corrected_scenarios.csv")

Error

error <- f_compute_error("2021-12-01", "2022-04-01", temp)

f_graph_error(
  error,
  "2022-01-07", 250 #publication date label
  )

#high VE all
# scenario <- read.csv("data/2022_01_07/beds_SC_peu_probable_low_VE.csv", sep=";") %>%
#   mutate(date = as.Date(date, format = "%d/%m/%Y", optional = T)) %>%
#   gather(key=scenarios, value = value, -date)
# scenario %>%
#   ggplot(aes(date, value, group=scenarios)) +
#   geom_line(data= scenario, 
#             aes(x=date, value, group=scenarios), color = "grey", size=1) +
#   geom_line(data= true_data_beds_hosp_rea, 
#             aes(x=date, y=rea, color = "red", group="réalité"), size=1) +
#   annotate('text', x = as.Date("2022-01-20"), y = 2800, label = "reality", color = "red", fontface = "bold",
#            family = "Times New Roman") +
#   annotate('text', x = as.Date("2021-12-25"), y = 5800, label = "scenarios", color = "grey", fontface = "bold",
#            family = "Times New Roman") +
#   xlim(date("2021-12-01"), date("2022-04-01")) + ylim(0,11000) +
#   labs(title = "Intensive Care Unit Beds",
#        subtitle = "scenarios published by the Pasteur Institute on January 7, 2022\n",
#        caption = "\ntwitter : @Covid_Scenarios")
# 
# scenario <- read.csv("data/2022_01_07/beds_SC_peu_probable_low_VE.csv", sep=";") %>%
#   mutate(date = as.Date(date, format = "%d/%m/%Y", optional = T))
# test <- f_compute_error("2021-12-01", "2022-04-01", "rea", true_data_beds_hosp_rea)
# ggplot(test, aes(date)) + 
#   geom_line(aes(y=error_med)) +
#   geom_ribbon(aes(ymin = error_min, ymax = error_max), alpha = 0.1) + 
#   geom_hline(yintercept = 100)

Winter 2021 curfew discussion

January 28 lockdown scenario

j’ai pris que confinement “light” 8 février. OK de ne pas prendre confinement “dur” mais je pense qu’il serait plus legit de regarder confinement qui commence le 1er février => extraire 3B

lockdown (“soft”, i.e. as in November 2020)starting on February 8th

#Admissions en réanimation
scenario <- read_csv("data/appendix/2021_01_28/2021_01_28.csv") %>%
  mutate(date = as.Date(date, format = "%d/%m/%Y", optional = T)) %>%
  select(-real)
scenario %>%
  ggplot(aes(date)) +
  geom_line(data= true_data_new_hosp_rea, 
            aes(x=date, y=new_rea_center), size = 1, color="red") +
  geom_line(aes(y=lockdown_light_Feb_8), color="grey") +
  geom_ribbon(aes(ymin = min_lockdown_light_Feb_8, ymax = max_lockdown_light_Feb_8), alpha = 0.4, fill="grey") +
  annotate(
    'text', x = as.Date("2021-01-28"), y = 170, 
    label = "reality\nno lockdown but 6 p.m. curfew", 
    color = "red", fontface = "bold",
    family = "Times New Roman") + 
  annotate(
    'text', x = as.Date("2021-02-15"), 
    y = 570, label = "lockdown scenario", 
    color = "grey", fontface = "bold",
    family = "Times New Roman") + 
  xlim(as.Date("2021-01-01"), as.Date("2021-03-15")) + g_theme +
  labs(title = "Intensive Care Unit Admissions",
       subtitle = "Pasteur Institute scenarios published on January 28, 2021",
       caption = "\ntwitter : @Covid_Scenarios")

lockdown (“soft”, i.e. as in November 2020)starting on February 1st

#Admissions en réanimation
scenario <- read_csv("data/appendix/2021_01_28/2021_01_28_lock_soft_Feb_1.csv") %>%
  mutate(date = as.Date(date, format = "%d/%m/%Y", optional = T)) %>%
  select(-real)
scenario %>%
  ggplot(aes(date)) +
  geom_line(data= true_data_new_hosp_rea, 
            aes(x=date, y=new_rea_center), size = 1, color="red") +
  geom_line(aes(y=lock_soft_Feb_1), color="grey") +
  geom_ribbon(aes(ymin = min_lock_soft_Feb_1, ymax = max_lock_soft_Feb_1), alpha = 0.4, fill="grey") +
  annotate(
    'text', x = as.Date("2021-02-15"), y = 170, 
    label = "reality\nno lockdown but 6 p.m. curfew", 
    color = "red", fontface = "bold",
    family = "Times New Roman") + 
  annotate(
    'text', x = as.Date("2021-02-15"), 
    y = 500, label = "lockdown scenario", 
    color = "grey", fontface = "bold",
    family = "Times New Roman") + 
  xlim(as.Date("2021-01-28"), as.Date("2021-03-08")) + g_theme +
  labs(title = "Intensive Care Unit Admissions",
       subtitle = "Pasteur Institute scenarios published on January 28, 2021",
       caption = "\ntwitter : @Covid_Scenarios")

February 14 scenarios

# #14 février
# scenario <- read.csv("data/2021_02_14/weekly_hospital.csv", sep=";") %>%
#   mutate(date = as.Date(date, format = "%d/%m/%Y", optional = T))
# scenario %>%
#   gather(key=scenario, value = value, -date) %>%
#   ggplot(aes(date, value+1700, group=scenario, color="scénarios")) + geom_smooth(se=F) +
#   geom_line(data= true_data_new_hosp_rea_weekly, 
#             aes(x=date, y=new_hosp_week, color = "réalité", group="réalité"), size = 1) +
#   annotate('text', x = as.Date("2021-01-05"), y = 7000, label = "reality", color = "red", fontface = "bold",
#            family = "Times New Roman") + 
#   annotate('text', x = as.Date("2021-02-27"), y = 25000, label = "scenarios", color = "grey", fontface = "bold",
#            family = "Times New Roman") + 
#   #confinement de 16 départements le 20 mars cf https://fr.wikipedia.org/wiki/Chronologie_de_la_pand%C3%A9mie_de_Covid-19_en_France
#   #2 semaines pour voir les effets sur hospitalisations. correspond aussi au confinement général du 3 avril
#   xlim(date("2021-01-01"), date("2021-03-27")) + ylim(0,37000) + g_theme +
#   labs(title = "Weekly number of hospital admissions",
#        subtitle = "scenarios published by INSERM on February 14, 2021\n",
#        caption = "\ntwitter : @Covid_Scenarios")
# 
# scenario <- scenario %>% select(-calibration)
# 
# test <- f_compute_error("2021-01-01", "2021-03-27", "new_hosp_week", true_data_new_hosp_rea_weekly)
# ggplot(test, aes(date)) + 
#   geom_line(aes(y=error_med)) +
#   geom_ribbon(aes(ymin = error_min, ymax = error_max), alpha = 0.1) + 
#   geom_hline(yintercept = 100)

Pasteur’s self-assessment

April 26, 2021 (assesses February 8, 2021 report)

June 16, 2021 (assesses May 21, 2021 report)

February 15, 2022 (assesses January 7, 2021 report)

Other non included (does not assesses hospitalizations)